The goal of this project is to develop a statistical tool to shed light on how the time course of treatment outcome is affected by medications, placebos, and other relevant factors. It addresses the RFA goal to Develop and apply advanced longitudinal statistical techniques to describe how the placebo response changes over the course of alcohol clinical trials and explore sources of heterogeneity in these response trajectories. Standard statistical methods such as linear modeling are not well suited for studying the very dynamic processes that affect outcome during and after treatment. Based on theoretically-justified assumptions about the time course of both medication and placebo effects, we propose a dynamic nonlinear statistical model. This model uses medication compliance and other time varying information to predict drinking during and after treatment. This model was applied to a random half-sample of data from Project COMBINE, a large multi-site alcohol treatment study. A linear model based on the COMBINE primary outcome analyses accounted for 6 percent of the variance of drinking over time; the nonlinear model was able to account for 87 percent. The nonlinear model requires further development to improve its fit to the time course of outcome, and to incorporate covariates. It will then be validated on the other half-sample of the COMBINE data. Furthermore, it will be validated again using data from Project Predict, a study parallel to Project COMBINE that was conducted in Germany. Modeling the time course of outcome, while technically challenging, makes it possible to differentiate between medication, placebo, and other important effects on outcome. This in turn makes it possible to ascertain more clearly where medications succeed as well as where they falter. The information from the tool developed in this study will help us to engineer better intervention packages to improve the health of persons with addictive disorders.